AD-Assurance-Lab/lab-drives

WMU AD Assurance Lab drives, 2019-2025

Smaller recordings from the AD Assurance Lab at Western Michigan University: a six-camera urban drive, radar test drives, radar-only drives, ZED 2i campus drives, a 2023 Basler camera test, two camera-LiDAR calibration recordings, a 2025 campus shuttle drive, a 2019 2D-lidar weather test and phone captures of a 2020 snowy highway drive. The Ouster point clouds from the same recordings are in a companion dataset, AD-Assurance-Lab/ouster-lidar-drives.

Collection	When	What	Size
kia_urban_2023-12	2023-12-04	one 27-minute urban drive, six cameras, GNSS, road weather, CAN	28.1 GB
radar_tests_2023-24	2023-12-11 to 2024-03-27	eight short radar test bags, radar, one camera, CAN	259 MB
radar_lidar_2020-21	2020-03-05, 2021-01-07	a highway and a city drive, radar only	14 MB
zed2i_campus_2022-07	2022-07-27, 2022-07-29	two campus drives, ZED 2i left camera	1.6 GB
lidar_weather_tests_2019	Oct-Nov 2019	13 static 2D-lidar runs in four weather conditions	2.5 MB
phone_snow_2020-02	2020-02-09	4 phone videos and 16 photos through the windshield on snowy highways	1.0 GB
basler_campus_2023-10	2023-10-16	two 40-second camera tests on a campus path and service road, one camera	172 MB
calibration_2023-24	2023-12-21, 2024-05-10	two camera-LiDAR chessboard calibration recordings, one camera each	959 MB
wmu_shuttle_2025-01	2025-01-29	one 3-minute campus shuttle drive, one camera	531 MB

Sensors

Sensor	Make and model	Collections	Source
Cameras	Basler, 1440×1080 (six at about 15 fps in the urban drive, one in the radar tests, one at about 10 fps on the shuttle, one in each calibration recording; 1280×720 at 30 fps in the 2023 campus test)	kia_urban_2023-12, radar_tests_2023-24, basler_campus_2023-10, calibration_2023-24, wmu_shuttle_2025-01	camera_info in the bags
Stereo camera	Stereolabs ZED 2i (left eye released), 1920×1080, about 14.6 fps	zed2i_campus_2022-07	Carow et al. 2023
Radar	Aptiv (Delphi) ESR 2.5	radar_tests_2023-24, radar_lidar_2020-21	Brown et al. 2022
Second radar	smartmicro UMRR-96 Type 153 (inferred)	point clouds in three radar_tests_2023-24 bags (2024-02-15, 2024-03-27)	Brown et al. 2023 sensor list
GNSS	Swift Navigation Duro (a ruggedized Piksi Multi)	kia_urban_2023-12, radar_tests_2023-24	topic names, Brown et al. 2022
Road weather	Lufft MARWIS (serial 10722)	kia_urban_2023-12	/LastValues topic
2D lidar	Slamtec RPLIDAR A2	lidar_weather_tests_2019	Goberville et al. 2020
Drive by wire	PolySync OSCC on the lab Kia Niro	kia_urban_2023-12, radar_tests_2023-24	Brown et al. 2022
LiDAR	Ouster OS1-64 Gen1, serial 991951000513 (IMU here, plus its metadata in the radar tests and the shuttle; point clouds in the companion dataset)	kia_urban_2023-12, radar_tests_2023-24, calibration_2023-24, wmu_shuttle_2025-01	sensor metadata

Conventions

Every camera stream comes with NAME.frames.parquet, giving per-frame bag time, header stamp and an MD5 of the original image bytes. Raw-image cameras are released as HEVC 4:4:4 at constant QP 13; JPEG cameras keep their original JPEGs, remuxed into MJPEG .mkv without re-encoding and verified frame for frame against the MD5s. Lossless FFV1 copies of the raw-image cameras are kept in the lab archive and are available on request. Other topics are in sensors/<topic>.parquet, one row per message, and each collection has a SHA256SUMS.

The vehicle CAN bus (/can_frame) is released whole, every ID logged. Only four IDs appear in any recording: 0x2B0 steering angle and 0x220 brake pressure (documented by OSCC and opendbc), and 0xB9 and 0x4B0, which are raw and undecoded. No OSCC module traffic was logged. can_scan.json lists every ID seen with its frame count.

kia_urban_2023-12: six-camera urban drive on the lab Kia Niro

One 27-minute drive on 2023-12-04 (1,612.6 s).

• Cameras: six Basler cameras at about 15 fps, 24,189 to 24,190 frames each, the raw camera JPEGs remuxed unchanged into camera_0N.mjpeg.mkv, with per-camera camera_info. The rectified streams stay in the archive bag.
• GNSS: 1,725 Piksi fixes at about 7 Hz, nearly all RTK, but only in the first 4.3 minutes of the drive, plus the Piksi IMU.
• Ouster IMU and /tf.
• Road condition: MARWIS readings (dry, wet or damp).
• Speed: the roscco can_frame_speed channel.
• CAN: /can_frame has every ID seen (0xB9, 0x220, 0x2B0, 0x4B0).

radar_tests_2023-24: radar and camera test bags on the lab Kia

Eight bags of 15 to 37 s each, recorded 2023-12-11 to 2024-03-27, 4.0 minutes in total.

Data	Bags	Form
Aptiv ESR tracks, status and raw radar-bus frames	2023-12-11 and 2024-01-31	sensors/radar_1_*
Second radar's point clouds	2024-02-15 and 2024-03-27	sensors/radar_*: x, y, z, radial speed, RCS and SNR per point
One Basler camera (raw RGB)	6 of the bags	HEVC (lossless FFV1 in the archive), 1,787 frames
roscco steering, brake and throttle reports, and the derived angle and speed channels	2024-01-31 and 2024-02-15	sensors/
Piksi fixes	2024-02-15 only: 278 standalone (SPP) fixes	sensors/

The raw bags of the four 2024-01-31 recordings are also public in AD-Assurance-Lab/infrastructure-sensing-doe-eems116 (radar retro-reflector tests, DOE award DE-EE0009657).

radar_lidar_2020-21: radar-only drives on the lab car

• 2020-03-05-radar_lidar_highway (134.4 s, highway): Aptiv ESR tracks, objects, status and the raw radar-bus frames (/can_tx, /can_rx). Its Ouster point clouds are in the companion dataset (highway_2020).
• 2021-01-07-radar_city (40.9 s, city): Aptiv ESR objects only (sensors/as_tx_objects.parquet).

zed2i_campus_2022-07: ZED 2i summer campus drives

Drive	Date	Duration	Frames
Engineering_Drive	2022-07-27	197.8 s	2,887
Campus_Drive_partial	2022-07-29	118.8 s	1,753

Both are ZED 2i left-camera colour at 1920×1080 and about 14.6 fps, with no GNSS, IMU or right eye. Campus_Drive_partial is the readable start of a recording whose file is damaged after 14.5 GB. The Trimble survey of the campus route (2022-07-28) is in the lab archive.

lidar_weather_tests_2019: 2D lidar in four weather conditions

Thirteen static runs with a 2D lidar (/scan) aimed at a target at 5, 10 and 15 ft, in rain, cloudy, snowy and fair conditions, October-November 2019, including one failed fair run that is flagged as such. runs.parquet gives, per run, the condition, distance, duration and scan count, the mean number of finite returns per scan, and the authors' published "average hits" on the target (a different measure).

basler_campus_2023-10: Basler camera test on campus

Two recordings on 2023-10-16 (39.8 s and 40.4 s, 1,201 and 1,211 frames) from a single Basler camera at 1280×720 and about 30 fps, mounted low and looking along a campus path and a curved service road, most likely at WMU. Only the camera was logged: no GNSS, IMU or vehicle data, and the platform is not recorded. HEVC, with the lossless FFV1 in the archive.

calibration_2023-24: camera-LiDAR calibration recordings

Two recordings of a chessboard target moved in front of a Basler camera and the Ouster, for camera-LiDAR calibration.

Recording	Where	Duration	Camera frames	Also
2023-12-21_cam_lidar	indoors, in the lab	91.8 s	354 at about 4 fps (camera_01)	camera_info, /tf
2024-05-10_chessboard	outdoors, a parking lot	137.7 s	4,130 at about 30 fps (camera_front)	camera_info, /tf, Ouster IMU

The cameras are HEVC (lossless FFV1 in the archive), with faces blurred (see Privacy). No calibration result was recorded: /tf holds only the Ouster's internal frames. The LiDAR scans are in the companion dataset (calibration_2023-24).

phone_snow_2020-02: snowy highway, phone captures

Four videos (1920×1080 at 30 fps, 6.0 minutes in total) and 16 photos (4032×3024) taken with a phone through a car's windshield on 2020-02-09, on snow-covered highways near Kalkaska in northern Michigan. The video streams and the photo image data are the phone's originals; the audio tracks and all file metadata were removed. captures.parquet keeps each file's UTC time, video duration and the phone's GPS position. The photos' positions are the phone's last fix and can lag by a few minutes.

wmu_shuttle_2025-01: campus shuttle drive

One 179.5 s campus drive on 2025-01-29: a Basler camera at 1440×1080 and about 10 fps, 1,724 frames, as HEVC (lossless FFV1 in the archive), with its camera_info, the Ouster IMU and metadata, and /tf. GNSS and CAN were not recorded. The Ouster point clouds and the lidar map built from this drive are in the companion dataset (shuttle_2025-01).

Privacy

The driving cameras face outward. Pedestrians are small and distant; parked-car plates are legible in a few frames and are not blurred, as in the lab's winter-driving release.

In calibration_2023-24 the people moving the chessboard have their faces blurred. Faces and heads were found with OpenCV YuNet and torchvision Keypoint R-CNN, each box was held two frames either side, and the frames left unblurred were checked by eye. In a few frames of 2023-12-21_cam_lidar the blur also covers part of the board. The MD5s in frames.parquet are of the original images, so they do not match these videos; the originals stay in the lab archive.

Citation

CC BY 4.0. Please cite the paper for the data you use:

• lidar_weather_tests_2019: N. A. Goberville, M. El-Yabroudi, M. Omwanas, J. Rojas, R. Meyer, Z. Asher and I. Abdel-Qader, "Analysis of LiDAR and Camera Data in Real-World Weather Conditions for Autonomous Vehicle Operations," SAE Technical Paper 2020-01-0093, 2020, doi:10.4271/2020-01-0093.
• zed2i_campus_2022-07: K. Carow, P. Kadav, J. Fanas Rojas and Z. Asher, "Projecting Lane Lines from Proxy High-Definition Maps for Automated Vehicle Perception in Road Occlusion Scenarios," SAE Technical Paper 2023-01-0051, 2023, doi:10.4271/2023-01-0051 (the paper's campus route; Engineering_Drive is not in the paper).
• radar_lidar_2020-21 and the lab Kia Niro platform: N. E. Brown et al., "Development of an Energy Efficient and Cost Effective Autonomous Vehicle Research Platform," Sensors 22(16):5999, 2022, doi:10.3390/s22165999. Related: N. E. Brown et al., "Higher Accuracy and Lower Computational Perception Environment Based Upon a Real-time Dynamic Region of Interest," SAE Technical Paper 2022-01-0078, doi:10.4271/2022-01-0078.
• radar_tests_2023-24: related to N. E. Brown et al., "Real World Use Case Evaluation of Radar Retro-reflectors for Autonomous Vehicle Lane Detection Applications," SAE Technical Paper 2024-01-2042, doi:10.4271/2024-01-2042.
• kia_urban_2023-12, basler_campus_2023-10, calibration_2023-24, phone_snow_2020-02, wmu_shuttle_2025-01: no paper yet; please cite this dataset.

@misc{wmu_lab_drives_2026,
  title        = {{WMU AD Assurance Lab} drives, 2019-2025},
  author       = {{AD Assurance Lab, Western Michigan University} and Asher, Zachary D.},
  year         = {2026},
  howpublished = {Hugging Face dataset},
  url          = {https://huggingface.co/datasets/AD-Assurance-Lab/lab-drives}
}

Contact: Zachary D. Asher, Western Michigan University, zach.asher@wmich.edu.